AMD Gets CUDA Compiler, Going All-In on HPC in 2016-2017

[Prysin]

1) It applies to all the code. My university actually discussed this with Nvidia representatives since we're getting a Xeon Phi node.

 

2) Google and Apple both make money. They license those apps.

are you sure said Nvidia rep didnt just tell you that to make sure the Uni didnt breed a whole generation of "loop-hole abusers"`??

[Guest Kloaked]

My take from this thread: no one is ever happy.

[Prysin]

My take from this thread: no one is ever happy.

no no no no, we are happy.

 

Both for AMD embracing CUDA. and because we can argue with Patrick

[patrickjp93]

are you sure said Nvidia rep didnt just tell you that to make sure the Uni didnt breed a whole generation of "loop-hole abusers"`??

While possible, it's doubtful. It would be the same principle they used on the HPC crowd. Nvidia would hate to have AMD steal its secret sauce.

 

no no no no, we are happy.

 

Both for AMD embracing CUDA. and because we can argue with Patrick

Argue? This is discussion. It's arguing when Svetlio is trying to skewer me b/c he can't accept the fact a college kid completely outclasses him at his own game.

[Prysin]

While possible, it's doubtful. It would be the same principle they used on the HPC crowd. Nvidia would hate to have AMD steal its secret sauce.

 

Argue? This is discussion. It's arguing when Svetlio is trying to skewer me b/c he can't accept the fact a college kid completely outclasses him at his own game.

well, like i said earlier... it is harder to tighten holes in IPs (legally, and get it approved) after the patent is filed. Then it is to open the patent to the public.

So if Nvidia cannot be bothered to go through a long arduous process to lock all the loopholes, then the easiest and cheapest way to solve the whole issue is to just say "nuh uh, you can't do that because...... we own it, yeah because we own it!"...

 

Aslong as AMDs hardware isnt directly reading RAW CUDA code, they may just be in a gray area aslong as the software is inbetween.

[patrickjp93]

well, like i said earlier... it is harder to tighten holes in IPs (legally, and get it approved) after the patent is filed. Then it is to open the patent to the public.

So if Nvidia cannot be bothered to go through a long arduous process to lock all the loopholes, then the easiest and cheapest way to solve the whole issue is to just say "nuh uh, you can't do that because...... we own it, yeah because we own it!"...

 

Aslong as AMDs hardware isnt directly reading RAW CUDA code, they may just be in a gray area aslong as the software is inbetween.

ISO got in trouble with Google about a GO cross-compiler as well. I wouldn't doubt Nvidia's ability to lock down its own IP when it's been making a walled garden with CUDA for so long.

[Prysin]

ISO got in trouble with Google about a GO cross-compiler as well. I wouldn't doubt Nvidia's ability to lock down its own IP when it's been making a walled garden with CUDA for so long.

you can build a wall, but all it takes to break in is some shoddy masonry or a cracked brick...

 

that or if you forget the spikes ontop of the wall... cuz then you can just jump over

[LAwLz]

royalties. If there is a sale and the original distributor is X, then X has claim to royalties unless the creator itself has formed a contract with the second party to sell via their stores.

But since we are talking about porting apps over, then royalties for sure.

 

Everything is about money. If apple and google didnt make any form of revenue out of this ordeal they would sue MS a long time ago

So Apple will get money for Facebook's app on Windows Phone? According to fairly well founded rumors the new Facebook app for WP will use project Islandwood. Does that money come from Microsoft or from Facebook?

Gonna a [Citation Needed] for that claim since I find it very strange that such a deal would happen.

 

And I don't think they have the legal ground to sue. The developers own the code and can do whatever they want with it. Apple does not own the source code for the Facebook app, and the only thing Microsoft's compiler does is map iOS APIs to WP APIs.

 

 

 

 

 

1) It applies to all the code. My university actually discussed this with Nvidia representatives since we're getting a Xeon Phi node.

[Citation Needed] on it applying to all code. The agreement you linked to only specifies it for the SDK and does not mention code you write yourself.

And no vague stuff like "it's implied" or "I personally talked to them". Those are not valid sources.

 

 

2) Google and Apple both make money. They license those apps.

[Citation Needed], like with Prysin.

[patrickjp93]

So Apple will get money for Facebook's app on Windows Phone? According to fairly well founded rumors the new Facebook app for WP will use project Islandwood. Does that money come from Microsoft or from Facebook?

Gonna a [Citation Needed] for that claim.

 

And I don't think they have the legal ground to sue. The developers own the code and can do whatever they want with it. Apple does not own the source code for the Facebook app, and the only thing Microsoft's compiler does is map iOS APIs to WP APIs.

 

 

 

 

 

[Citation Needed] on it applying to all code. The agreement you linked to only specifies it for the SDK and does not mention code you write yourself.

And no vague stuff like "it's implied" or "I personally talked to them". Those are not valid sources.

 

 

[Citation Needed], like with Prysin.

If ported originally from an Apple Phone, yes, because that's how the apple developer license agreement works.

 

No it doesn't. You just read it the way you want to.

 

2) See Apple developer license agreement and android developer license agreement.

[LAwLz]

If ported originally from an Apple Phone, yes, because that's how the apple developer license agreement works.

[Citation Needed]

If you ask Microsoft then it's not a port either. It's just there to "help you write great Windows apps that use as much of your existing code and knowledge as possible".

I have not found anything about developers or Microsoft having to pay Apple if they reuse code from their iOS apps either. Such a big drawback would most likely be clearly outlined if it existed.

 

 

No it doesn't. You just read it the way you want to.

No I am actually reading it exactly as it was written. I have not read the entire agreement so it is possible that I have missed it, but so far I have only been able to find things that apply to the SDK itself, not the code you write. I have asked you to link me to the part of the agreement that says the code you write targeting CUDA is not to be reused in other programs and so far you have not delivered. If it applies to the code you write then it will be specified in the agreement. If it is not specified in the agreement then it does not apply. It's as simple as that.

[patrickjp93]

No I am actually reading it exactly as it was written. I have not read the entire agreement so it is possible that I have missed it, but so far I have only been able to find things that apply to the SDK itself, not the code you write. I have asked you to link me to the part of the agreement that says the code you write targeting CUDA is not to be reused in other programs and so far you have not delivered. If it applies to the code you write then it will be specified in the agreement. If it is not specified in the agreement then it does not apply. It's as simple as that.

It's exactly the statements you've quoted. They apply to everything you write since, omg, it's technically part of the CUDA SDK since, dear god, you used it to build your code and you have to abide by the terms of usage. 

 

My capstone is building an Apple Watch application. If we use that code to translate the functionality over to an Android platform, we owe royalties to apple on all sales of the android version of the app as part of the agreement you sign when you get a developer license and publish to the store. While I'm not in a position to go digging through that contract right now (on my phone), it's there. I'm sorry but you're being willfully ignorant in both of these cases.

[Notional]

Does it mean that AMD cards could suck less at nvidia gameworks in the future?

 

Not at all. This is a tool and help for developers to port CUDA programming into their HIP compiler for AMD and HSA. As pretty much all developers doesn't even have access to the source code of GameWorks, there is no way they can or are allowed to recompile it to anything else. We are also talking compute, not graphics. I doubt any dev is allowed to recompile APEX code to anything else.

 

1) It applies to all the code. My university actually discussed this with Nvidia representatives since we're getting a Xeon Phi node.

 

2) Google and Apple both make money. They license those apps.

 

No way in hell NVidia owns any programming made by anyone in CUDA. That is covered by IP rights of the individual programmer/company, just like music is. Even IF Nvidia made such a bonkers EULA, it would crash and burn in pretty much any courtroom. EULA/contracts < Law. I can guarantee that such a EULA is completely void in EU countries.

[patrickjp93]

Not at all. This is a tool and help for developers to port CUDA programming into their HIP compiler for AMD and HSA. As pretty much all developers doesn't even have access to the source code of GameWorks, there is no way they can or are allowed to recompile it to anything else. We are also talking compute, not graphics. I doubt any dev is allowed to recompile APEX code to anything else.

 

 

No way in hell NVidia owns any programming made by anyone in CUDA. That is covered by IP rights of the individual programmer/company, just like music is. Even IF Nvidia made such a bonkers EULA, it would crash and burn in pretty much any courtroom. EULA/contracts < Law. I can guarantee that such a EULA is completely void in EU countries.

Nvidia owns the rights to that program's translation in commercial software. If you decide to write something in a proprietary language, sorry, but you can be bound under some less than savory rules.

 

The programmer only owns the algorithm/overall design (the pseudo code). Nvidia owns the syntax and all manipulations thereof.

 

Contracts are law. There's an entire branch of law dedicated to contracts, and sorry but it satisfies all 4 requirements: consent, competency, and the 2-part quid pro quo.

 

I can guarantee you it's not.

[niofalpha]

Now what will people blame on AMD cards under performing in Gameworks games on? /s

 

So CUDA is basically the main language in the HPC space?

[TechStyle]

Hm... one less reason to stay on green team. Premiere will enjoy CUDA on AMD.

[Prysin]

Now what will people blame on AMD cards under performing in Gameworks games on? /s

 

So CUDA is basically the main language in the HPC space?

CUDA has nothing to do with gameworks per say. This wont change a damn thing with gameworks performance what so fucking ever.

 

Gameworks leverages TESSELATION. Which CUDA wont help with. Because tesselation performance is based in GPU architecture. and AMDs architecture SUCKS AT TESSELATION.

[niofalpha]

CUDA has nothing to do with gameworks per say. This wont change a damn thing with gameworks performance what so fucking ever.

 

Gameworks leverages TESSELATION. Which CUDA wont help with. Because tesselation performance is based in GPU architecture. and AMDs architecture SUCKS AT TESSELATION.

CUDA effects are based around game works tho...

It speeds up OptiX https://developer.nvidia.com/gameworks-optix-overview 

I know I saw something on the Nvidia site saying PhysX effects are CUDA accelerated. Finding link now.

This I think? Not the link I was looking for but should be suffice.

https://en.wikipedia.org/wiki/PhysX

[Altecice]

Server-Based GPU's

(Not home)

[Notional]

Nvidia owns the rights to that program's translation in commercial software. If you decide to write something in a proprietary language, sorry, but you can be bound under some less than savory rules.

 

The programmer only owns the algorithm/overall design (the pseudo code). Nvidia owns the syntax and all manipulations thereof.

 

Contracts are law. There's an entire branch of law dedicated to contracts, and sorry but it satisfies all 4 requirements: consent, competency, and the 2-part quid pro quo.

 

I can guarantee you it's not.

 

In Danish law, we have a term called Præceptiv or Ufravigelig. It basically means Unconditional or Mandatory. Some laws are still in function even with consent. You cannot consent to murder for instance. In EU, most consumer law is mandatory and you cannot consent to dismiss them. The same is usually true in IP rights/law as well. I don't know the details in American law on this matter, but America is not the entire world (despite what Americans themselves seems to think).

 

Right now you can only speculate that AMD has gotten a CUDA license. I personally doubt they have, but I have no proof either way, so I'm open for it. However it is very clear that AMD does not and will not support CUDA on their hardware what so ever. They merely have a "translator" for CUDA programming to their own HIP compiler. This is a great way for AMD to help developers get away from CUDA, not further implement CUDA into the market.

 

CUDA has nothing to do with gameworks per say. This wont change a damn thing with gameworks performance what so fucking ever.

 

Gameworks leverages TESSELATION. Which CUDA wont help with. Because tesselation performance is based in GPU architecture. and AMDs architecture SUCKS AT TESSELATION.

 

All APEX based GameWorks effects, utilizes CUDA, such as Turbulence in VisualFX and all PhysX based effects under Physics. Most VisualFX in GameWorks are tessellation based though. You can see more here: https://developer.nvidia.com/content/introducing-nvidia-gameworks 

[LAwLz]

It's exactly the statements you've quoted. They apply to everything you write since, omg, it's technically part of the CUDA SDK since, dear god, you used it to build your code and you have to abide by the terms of usage. 

No, the code I write is not part of the CUDA SDK. It's the same as the code I write not being part of Visual Studio, and therefore does not have to follow the same licensing deal as Visual Studio. By your logic, that the source code is "part of the SDK", also means that every single GPL program compiled in Visual Studio is breaking the law because that program should be licensed under the same license as Visual Studio, and not the GPL. Do you see how absurd that is?

The source code you write is NOT part of the SDK and does not have to follow the exact same license.

If I write a book in Microsoft Word then that book then I get the copyright for it, not Microsoft.

 

The only way for Nvidia to be able to do what you are saying, is if they had in the user agreement that they get copyright for all code written with the intention of using the CUDA SDK. I can't find anything like that in their agreement so I asked you where you read it. For Nvidia to have the authority over other peoples' copyrighted work it must be in the contract in plain English. Not some "it is implied" or "someone from Nvidia told me in person". So I will ask you again. Can you please cite the part of the agreement that states that Nvidia gets the copyright for the code you write?

 

 

My capstone is building an Apple Watch application. If we use that code to translate the functionality over to an Android platform, we owe royalties to apple on all sales of the android version of the app as part of the agreement you sign when you get a developer license and publish to the store. While I'm not in a position to go digging through that contract right now (on my phone), it's there. I'm sorry but you're being willfully ignorant in both of these cases.

[Citation Needed]

So if you write printf("hello world") in some iOS program then it is illegal for you to copy/paste that into some other program you write? That's laughable and I doubt that would hold up in any court.

[Peter Vaughan Truslow]

Where have you been? Native multithreading's been here since C++11.

 

Compiler flag -D_GLIBCXX_PARALLEL https://gcc.gnu.org/onlinedocs/libstdc++/manual/parallel_mode.html

I stand corrected, I'm fluent in C, but am Still learning C++

my professor hasn't taught anything exclusive to C++11 thanks of the info, I'll be reading up on that.

[Dabombinable]

Finally the idiots have accepted the third offer.

[patrickjp93]

Now what will people blame on AMD cards under performing in Gameworks games on? /s

So CUDA is basically the main language in the HPC space?

HPC has been in the CUDA garden mainly for years. It's only with the Xeon Phi that OpenMP/OpenACC and OpenCL became remotely relevant.

[patrickjp93]

CUDA has nothing to do with gameworks per say. This wont change a damn thing with gameworks performance what so fucking ever.

Gameworks leverages TESSELATION. Which CUDA wont help with. Because tesselation performance is based in GPU architecture. and AMDs architecture SUCKS AT TESSELATION.

Much of the PhysX work is CUDA-based.

[Guest]

BS. Nvidia just doesn't focus on it because has CUDA and a fairly good ecosystem and marketshare. OpenCL 2.0 support is inbound.

 

Nvidia is the king of compute, and arguably CUDA is way ahead on compute technologies than OpenCL is. Asynchronous compute kernels was in CUDA long before it was in OpenCL (note this is different from doing graphics rendering and computing kernels at the same time asynchronously).

Well, the way you worded this, it makes me feel like your an nVidia fanboy.

 

Also, AMD helped develop OpenCL with Apple & Khronos Group.

They know how to implement OpenCL better than any other GPU Manufacturer.